My feeling is that lower requirement for table regions should be: my_table_region_count > REGION_SERVER_count*3.
Each Region server should get at least one table region, so your read/write load would be evenly distributed across all region servers in any cases. *Assumption is that your data is not skewed, you never ever have hot spots. If it's skewed, you would start to solve completely different problems :) 2015-09-07 14:59 GMT+02:00 Ted Yu <yuzhih...@gmail.com>: > For the 96 region table, region size is too small. > > In production, I have seen region size as high as 50GB. > > FYI > > > > > On Sep 7, 2015, at 2:55 AM, Akmal Abbasov <akmal.abba...@icloud.com> > wrote: > > > > Hi, > > I would like to know about pros and cons against small region sizes. > > Currently I have cluster with 5 nodes, which serve 5 tables, but there > are ~80 regions per node, while actual data(total size of all hstores) is > ~50GB. > > Isn’t it an overhead, since there is a table which is ~30MB which has 96 > regions. > > I was thinking about merging regions, because of overhead for managing > them(metadata, memstore per region, more flushes, more compactions). > > Any suggestions? What is the avg region size in your case? > > > > Thanks. > > > > >